We describe the Story Intention Graph, a set of discourse relations designed to represent aspects of narrative. Compared to prior models, ours is a novel synthesis of the notions of goal, plan, intention, outcome, affect and time that is amenable to corpus annotation. We describe a collection project, DramaBank, which includes encodings of texts ranging from small fables to epic poetry and contemporary nonfiction.

We propose a methodology for a novel type of discourse annotation whose model is tuned to the analysis of a text as narrative. This is intended to be the basis of a “story bank” resource that would facilitate the automatic analysis of narrative structure and content. The methodology calls for annotators to construct propositions that approximate a reference text, by selecting predicates and arguments from among controlled vocabularies drawn from resources such as WordNet and VerbNet. Annotators then integrate the propositions into a conceptual graph that maps out the entire discourse; the edges represent temporal, causal and other relationships at the level of story content. Because annotators must identify the recurring objects and themes that appear in the text, they also perform coreference resolution and word sense disambiguation as they encode propositions. We describe a collection experiment and a method for determining inter-annotator agreement when multiple annotators encode the same short story. Finally, we describe ongoing work toward extending the method to integrate the annotator’s interpretations of character agency (the goals, plans and beliefs that are relevant, yet not explictly stated in the text).

Digital image collections in libraries and other curatorial institutions grow too rapidly to create new descriptive metadata for subject matter search or browsing. CLiMB (Computational Linguistics for Metadata Building) was a project designed to address this dilemma that involved computer scientists, linguists, librarians, and art librarians. The CLiMB project followed an iterative evaluation model: each next phase of the project emerged from the results of an evaluation. After assembling a suite of text processing tools to be used in extracting metada, we conducted a formative evaluation with thirteen participants, using a survey in which we varied the order and type of four conditions under which respondents would propose or select image search terms. Results of the formative evaluation led us to conclude that a CLiMB ToolKit would work best if its main function was to propose terms for users to review. After implementing a prototype ToolKit using a browser interface, we conducted an evaluation with ten experts. Users found the ToolKit very habitable, remained consistently satisfied throughout a lengthy evaluation, and selected a large number of terms per image.